01 / AUTONOMY
Loss of Control
What happens when autonomous systems take actions operators did not intend, understand or successfully interrupt?
HUMANITY BEFORE UNCHECKED AUTONOMY
A SAFER FUTURE
STAYS HUMAN.
Advanced artificial intelligence could reshape security, work, privacy and human decision-making. This observatory continuously surfaces documented incidents, credible reporting and unresolved safety questions.
EXPLORE THE LATEST →REAL PEOPLE. REAL RISKS.INDEPENDENT REPORTING.A MORE HUMAN FUTURE IS STILL A CHOICE.
THE RISK LANDSCAPE
THE SYSTEMS
ARE GETTING
MORE CAPABLE.
Capability is moving quickly. Control, verification, accountability and governance have to keep up.
02 / SOCIETY
Mass Surveillance
AI can expand identification, tracking, inference and behavioral monitoring at unprecedented scale.
03 / SECURITY
Weapons & Cyber
Increasing autonomy can lower the threshold for cyber operations and compress decision time in high-stakes environments.
04 / SCALE
Concentrated Power
Vast compute and highly capable models can concentrate consequential technical power in a small number of institutions.
RISK
Loss of Control
More capable systems can take actions humans struggle to predict, audit or stop.
SOCIETY
Mass Surveillance
AI can amplify tracking, profiling and behavioral inference across populations.
GLOBAL SECURITY
Autonomous Weapons
Greater autonomy can compress decision time in conflict and reduce meaningful human control.
INFRASTRUCTURE
Concentration of Power
Massive compute can concentrate consequential technical power in very few institutions.
AUTOMATICALLY UPDATED
THE
INCIDENT WIRE.
Current reporting on unexpected AI behavior, cyber misuse, autonomous systems, safety failures and related concerns.
Connecting to current reporting…
LOADING CURRENT REPORTING…
FEED METHOD — Articles are selected automatically from current news feeds using safety- and incident-related search terms. Inclusion is not independent verification or endorsement. Open the original publisher for full reporting and context.
THE SCALE QUESTION
WHEN MONITORING
BECOMES AUTOMATIC.
Systems that can recognize, classify, infer and act across millions of signals change the scale of surveillance—and the consequences of error.
EPHEMERAL PUBLIC ROOM
LIVE
DISCUSSION.
Talk with other visitors about current reporting, safety research and the future of advanced AI.
24-HOUR VISIBLE HISTORY • NO ACCOUNT REQUIREDENTER THE ROOM
Choose a temporary name.
Do not share personal information. Threats, targeted harassment, spam and illegal content are prohibited.
WHAT WE KNOW / WHAT WE DON'T
RISK IS NOT
THE SAME AS
CERTAINTY.
There is no scientific consensus that advanced AI will cause human extinction.
There is also no established method that guarantees future highly capable systems will remain controllable. Researchers disagree about probabilities, timelines and the safeguards that would be sufficient.
INTERNATIONAL AI SAFETY REPORT ↗ AI SECURITY INSTITUTE INCIDENT REPORTING ↗ CENTER FOR AI SAFETY ↗UNCERTAINTY
IS NOT
A SAFETY CASE.
OPEN QUESTIONS
01
Can highly capable autonomous systems always be reliably interrupted?
02
Can independent evaluators verify a safety case before deployment?
03
Who is accountable when autonomous systems act outside intended boundaries?
04
What prevents competition from rewarding speed over caution?
05